Google I/O 2014 – How crashing my startup taught me to work at Y Combinator

name’s Aaron Harris. I’m a partner at Y Combinator. And I’m going to tell you a
little bit about my experience at YC and how crashing the
company that I started at YC has actually made me
better at being a partner, giving advice,
and helping people avoid the same fate
that we endured. So that’s a reasonably
decent picture of me. Ignore the watermark behind it. And that’s how you can
find me on Twitter. If anyone wants to
get in touch, that’s probably the best way to do it. So what is Y Combinator? I know a lot of people know. But the truth is a lot of
people are kind of confused about what we do. It’s actually pretty simple. YC is actually just
a big orange room where once a week we have
really interesting speakers– this is Mitch Kapor– come
and talk to our founders and tell them,
here’s what happened to me when I
started my start-up. Here are the things
that I learned. Here are the things
that I screwed up, did well, funny
stories, war stories, all these kinds of things. And it’s pretty great,
because sitting out there, especially as someone who
is a first-time founder, you have no idea
what’s going on. And at YC, our goal is
to take really, really amazing founders, whether
first time or repeat founders, and try to make them
better, try to give them advice, along with
some money, and improve their chances of success. So actually, this
was my YC batch. And that’s me right over there. I’m lucky I made it into
the frame of this shot, because I almost got
bumped off, and then I wouldn’t have a good visual. But this was me
in winter of 2011. So about four years ago I
went through YC Combinator with my company, Tutorspree. Now, when we first came up
with the idea for Tuturspree, we had really, really
bad design skills. And this is basically
what we put up. And we said, we’re going to make
an online system for connecting parents and tutors for
one-on-one learning. We’re like, this is
going to be amazing. No one’s ever tried
anything like this before. It’s revolutionary. It’s amazing. What’s the first step? Well, we’re going to send
out a lot of emails to people and just try to sign
up a bunch of tutors. And in order to do that,
we needed a sign-up form. And so we put this up. Now, I wouldn’t
call this a launch of any kind, because
maybe 30 people saw it. But we sent it out
to a couple people, and we didn’t make a big deal. We didn’t want to send it
out to everyone we ever knew, because we thought, meh, not
really ready to be a company, not ready for launch. We couldn’t set up any lessons. We couldn’t take payment. We didn’t have profiles. We didn’t really have anything. It was really just
a landing page. And so we went back
to the drawing board. We’re like, OK, we’ve got to
make this a little better. It’s got to be prettier. Let’s show everyone all
the tutors that we have. And maybe now we’re ready
to be a real company. Now we’re ready to go raise
money and do this thing and do that thing and
run people through it. But actually, this was
still just a landing page. And while it had
a little bit more, honestly, still not ready. But it was at this point
that we said, OK, we want to build this. We’re committing
to this full time. Let’s see what we can do. And so we went and
applied to Y Combinator. And we applied with just
the shell of the website. This is actually roughly
what we applied to YC with. Still not that great
looking and really, really basic functionality. Basically, if you filled
that little form out, if you entered an
address, it would show you a list of local tutors. And we could kind of
match you up with them, but we couldn’t take
any payment from you. We had maybe 20
profiles on the site. Really not much there. But we got into YC anyway,
because apparently we fooled them into thinking
that we were actually fairly credible and
fairly good and had some idea of what we were doing. And as soon as we got
there, PG said, OK, launch. No, no, no, we’re
not ready to launch. Launch site, it’s unpolished. It doesn’t do the thing. We don’t have full coverage
in all of our cities. They said, no, no, no, launch. Nah, we’ve got to build
all these new features. We know that people are going
to want fancy scheduling, launch this, launch that. And he kept
hammering away at us, and we said, ah, what the hell. Let’s just launch the damn
site and see what happened. And that was terrifying. That was kind of a weird moment,
because we didn’t have any idea whether or not anyone would
actually react to this thing. Would anyone actually
use it beyond the friends that we had forced
to create profiles? And on the day of our launch,
we got this really nice headline off TechCrunch that Tuturspree
was the Airbnb for tutoring, which was exactly our message. That was our pitch. Airbnb for tutoring,
that’s so cool. Airbnb was YC, we’re YC. They help people make money
off things that they have. We’re going to– it’s
going to be great. Our basic logic
boiled down to this. Airbnb is successful. We’re kind of like Airbnb,
therefore we’ll be successful. It was simple. It was honestly– it
was a full-proof plan. It couldn’t fail. And we even worked
out a really nice flow for how our customers
would interact with us. Parents or students
would get to the site. They’d search for a tutor. They’d find the tutor. They’d send that
tutor a message, and we would make money. The problem was it didn’t work. There was this thing
that we saw that started to happen where we definitely
had people starting to come to the site
and looking for tutors. We got that because we
had had some nice press. And they were actually
sort of finding tutors, because we were localized to
the Bay Area and to New York. And we said, if
you’re here, enter your information and a
tutor’s going to pop up. We were pushing through as many
tutors as we possibly could. And then we’re like, OK, now
the Airbnb thing happens, right? We grow, we get really
rich, everyone’s happy. Amazing. But there was something
funny happening between the finding the tutor
and the booking the lesson. We weren’t sure
what it was, but we started taking a look
at our messaging layer and actually
watching the message traffic go back and
forth between parents and tutors or
students and tutors and watch what was coming back. And this really
weird thing started happening that was completely
counter to what we believed before we had
launched Tutorspree. We had this perfect
platonic ideal of what our company would be,
of how the users would actually use it. And the real world brutally
reminded us that we were wrong. So here’s the things
that we learned. First of all, our
customers weren’t students, as we
initially believed. We thought, well, if the student
was matching up with the tutor, then that student
was the customer. But in reality, the parents
were the ones paying for things, and they where the customer,
which is a completely different set of user behaviors
and user models around what actually
people want and how they’re going to interact. And so the thing
that we were missing in this wide open marketplace
model where people were basically bidding
on tutors and tutors had all these crazy
prices and no one knew what to do with anything
was that we were lacking trust. And what we really
needed was the trust of not just the students,
not just the tutors, but of the parents. And we didn’t know
how to get that. We didn’t know why
they weren’t using it. We could kind of
see in the messaging that people were not
responding at quite the level that we needed them to,
but we had no idea why. And so the thing we needed to do
was we needed to talk to them. And we needed to
talk to them a lot. This is one of these
really unsexy things. It’s the thing
that you don’t want to think about as a founder. You want to think that, OK, I
built this brilliant product. I’m perfect, it’s
perfect, so therefore it’s going to do great. The users will do what
I tell them to do. There’s actually a
core message that we tell our startups at YC. And this is– this was like
me learning a lesson really well by running face first into
a brick wall, Which is probably the best way to learn
things, because otherwise you don’t really remember it. And the truth is
what you have to do is you have to get
the product out there, and then you have
to talk to the users who you think are going to use
it, who you want to use it. But more important
than the ones that you think are going to
use it, you have to talk to the people who are
actually using your product. Because if you haven’t talked
to the people who are actually using it, you won’t
understand the behaviors that they’re going
through while using it or their rationale
for wanting to use it. And so we started
doing this thing where we’re talking to them, talking
to them, talking to them. And it actually led us
to completely change the core model of Tutorspree. We realized that we weren’t
this wide-open marketplace. We were not this ideal of a
place where people magically find each other and agree to
pay one another for lessons. We were a system where we
actually had to hold hands. We had to work with each
parent and make sure that they were happy until
we really figured out how things were going. So we did that. And it worked
really, really well. In fact, in the six
months between when we started doing that–
well, within six months, we’d increased our
revenue dramatically. We had increased our profit
even more dramatically. And the thing that was
actually most important to us was we increased the
engagement of our customers by three times. That meant they were doing
three times more lessons. They were also getting to
the point of booking lessons much more quickly. They were much more satisfied. They sent us love notes instead
of hate mail, which was better. We were on the phones with
them, so they knew us. They saw us as a partner and
not just a random utility that they could run back and
forth, in from and away from. And we never would
have known that had we sat inside of our office,
which was just the living room of our apartment,
not putting a product out and had we not been
talking to the customers. And that’s the core
of Y Combinator. We actually tell our
companies that there’s really only two things
to do during YC. You write code and you
talk to your users. And by doing those
two things, you end up making
something people want. And at the end of
the day, a startup is about making
something people want. A company is about making
something people want. Now, there are
variations of this. You could say you make
something people need. But the truth is, if you make
something people need and they don’t want it, they’re probably
not going to use it that much . And certainly, they’re not going
to tell their friends about it. There aren’t a lot of
school kids– well, let’s say a lot
of kindergartners who are really excited
about broccoli. But they probably
need to eat broccoli. They don’t want to,
but they need to. And so if you can
get to want, if you can be the thing that
people really want, that’s when you start winning. And as we started
building this product that was working, as people
were starting to interact with it more, we
noticed that more people were showing up at the site. We hadn’t really thought enough
about why that was happening. And what we discovered was
there was a very good reason why it was happening. There were some very
intelligent people sitting in some
buildings in this campus, creating algorithms that were
sending people to our site. Google was where we
got our customers. And we were getting
our customers, because at that point
we had created something that customers actually wanted,
which Google saw and measured. Google saw people
hitting our site, sticking around, transacting,
and it worked out really, really well for us. And so let’s put that
aside for a second. We said, OK, well, SEO
is working really, really well for us– just
brilliantly good traffic. It’s basically free for us,
converts really well, amazing. What else should we
be thinking about? Well, we had ideas around
other things, right? There are other ways
to acquire customers. You’ve got display
advertising, Facebook, SEM. We started to do partnerships. We talked to– about
doing Rev shares. We talked about doing
affiliate marketing, giveaways, flyering communities. We talked to administrators
of schools, PTAs, colleges. We tried to get guest blogs. And everything we did paled
in comparison to the channel that we had. But that led us to do
something a little strange. Because as we were
focused on SEO, we actually started forgetting
about our customers. We started forgetting about
what it was they wanted and started thinking
only of what was going to drive more
traffic to the site. So we basically tried
to play games with SEO. We didn’t do anything
particularly shady. But what we did start
doing was things that were probably not what
customers actually wanted. And Google picked up on
that, because the algorithms are designed to send people
to the things that they want. And so on March 16th of
2013, our traffic fell by 80% overnight. It was a really crappy morning. Along with our traffic dropping,
our revenue fell off a cliff. Our customers couldn’t find us,
new customers couldn’t find us. And everything
basically went to hell. And we sat around, and we
kind of were tearing our hair and trying to figure
out, well, what was it that caused
this to happen? Hadn’t we done the right things? We had some customers
that liked us. They found us. They were using us,
so what was going on? So we started
trying to figure out why it was that Google
had decided that we were no longer making a thing as
good as our customers wanted, because they were
starting to bounce more. They were getting more confused. That’s when we started learning
about the algorithm updates. Now, we had known how
the algorithm worked when we started Tutorspree. We had paid a lot
of attention to it, and it dovetailed
perfectly with our strategy for making something that our
customers actually wanted. But the new
algorithm saw that we were doing things
that weren’t great. We had duplicate
content within our site, because we had tried to create
a lot of pages representing different categories. We hadn’t done it
in the best way. We had thin content on pages. So for instance, we had a
single math tutor in, let’s say, Wichita, Kansas. And for that one math
tutor, we had a page for math tutors, algebra tutors,
calculus tutors, calculus II tutors, and trigonometry tutors. That one tutor was showing
up on seven or eight pages and is the only
content on that page. That was bad. If a customer had hit that page
and then said, oh, well, let me see what else
you have, they were going to see the same
thing again and again, get frustrated, decide we
were full of crap, and leave. And that’s what
we were learning. Wrapped up in that was the
fact that our navigation wasn’t good enough. People couldn’t find
their way around the site. Yeah, they could get from the
home page to one other page. But we discovered
they were actually landing on deeper pages first
and then not understanding what they were going to have to
do to get to where they needed. And finally, the
freshness of our pages, the speed with which
we updated the pages and actually made things
look like somebody was home, fell off. And so looking at all that,
what we realized was we had gotten lazy. We had had a model
originally, in the first days of Tutorspree, of how
Tutorspree worked. And we brought
that out to market and discovered it was wrong. And that led to all
this great iteration and all this great building of
what people wanted us to do, which is like the
core message of YC. But then we were out of YC,
and we stopped paying attention to some of these signals. And as a result, we
started optimizing, not towards what
our customers wanted but what we’ve perceived
certain channels wanted us to do in order to
bring customers to us. So we took a step back, and
we redesigned everything about the site. We actually built the site to
be what our customers wanted, to model on their behaviors
of how they worked through the flow of our site,
how we held their hands, how we matched them up with tutors,
how we resolved conflicts with tutors or made them
happy when things went wrong. And in doing so, we
completely changed the architecture of the
site and what we had. We cut over a million
pages from the site. We made sure that every
single important page had fresh, good,
useful content that explained to our users exactly
what was on that page, what kind of value they’d get from
it, what they could expect to pay for tutors– all
kinds of good information that they actually found useful. At the same time, we eliminated
duplicate geographies. If someone was
looking for 90023, they should see the same thing
as 90024, or that location if they entered in words. In New York, say, 1023 and
1024, both the Upper West Side of Manhattan. It should show the same page. And as we started doing this,
we started seeing improvements. Our rankings on Google improved. Our better UX and better UI led
to our customers being happier. And overall, our
site was faster. It was friendlier. And quite frankly,
it was just straight up betterer, which is not a
word, for all of our customers. And it was great. It was starting to
work really well. But at the end of the day,
it didn’t help enough. When you try to get
to Tutorspree now, you just get a 404. The reason was we had fallen
too far behind in that process. Startups basically
are constantly fighting, in my mind,
against two things. They’re fighting against
apathy, because no one knows who you are. No one cares who you are. And they’re fighting
against time. You only have a certain
amount of time to do things. And that’s whether you run
out of money or the people who work for you get bored
or the customers who thought you had decide
you’re not doing the things that they want. Any of those things
can kill you. And for us, what
it came down to was we realized that
we just could not put enough people onto
the site in the way that we thought
was best to keep it as a rapidly scaling business. It was pretty tough
to understand, because we had gotten lazy. We have forgotten the
right things to do. And by doing that, we had
doomed our own startup. And that’s kind of
a tough thing to say when you spend
two or three years of your life on something. But what it does is it
makes it so that when I talk to our new companies,
to new companies in the batch, I can tell them
how important it is to just get their
product out there and to iterate– and not only
that, never stop iterating. Launching doesn’t just
mean getting your site up on the internet
for the first time or getting your first press hit. It means constantly launching
things for your users that they want to use, that
they love, that make them happy. Because if you do
that, you actually make something that they want. At the end of the day, that’s
the only thing that counts. So if you ever find
yourself drifting away from what people
want– and you can track that in a million ways. You can use Google
Analytics to see how people are interacting
with your site. You can watch your revenue,
you can watch your user growth, engagement, all these things. If you find that
you’re moving away from what they want, and
you’re honest about that, then you just repeat
the first step. You figure out what they
want, you talk to them, and you launch new things. And that’s kind of
the message that I try to give to every
single one of our founders. And hopefully it’ll
sink in better for them than it did for me. That’s it. Thank you. [APPLAUSE] AARON HARRIS: What? Oh, thanks. I think I have about
10 minutes left. If anyone has any questions,
I’d be happy to answer them. I also have two of my partners. Kevin Hale is back there and
[INAUDIBLE] is over there, so they can also–
I’m volunteering them to answer
questions as well. If anyone has specific
questions about YC afterwards, we’ll be hanging around. So any questions? Yeah. AUDIENCE: [INAUDIBLE]. AARON HARRIS: Yeah. AUDIENCE: [INAUDIBLE]. AARON HARRIS: Yeah,
so the question was, when our traffic dropped
off, why did it do that? Didn’t people still need tutors? And so the reason that we–
our core hypothesis on this is that it’s an incredibly
fragmented market. Actually, most of
the tutoring goes to these little mom and
pop shops that are local. And that’s what we
were fighting with. So we had to create
a better experience than local recommendation
from one person to another. That’s really, really hard. And so as soon as we
started falling below that, we lost a lot of
our customer base. And once your reputation
starts getting tarnished, it’s really, really
hard to win it back. Parents don’t give
you a second chance. Anyone else? Yeah. AUDIENCE: [INAUDIBLE]? AARON HARRIS: Yeah,
so the question was, was there a specific
set of metrics or were there specific
metrics that convinced us that it was– we were done? It wasn’t any one metric. It was sort of a cumulative
view of everything that was going on, looking
at what our user acquisition cost was doing relative to
our value of a customer, comparing that with
where we needed it to be to rapidly scale
through multiple channels. And we realized that
those things just were no longer going to
work out the way that we thought they would. We couldn’t figure out how
to really make them work. So it’s kind of
just– at some point you hit this confidence interval
that– alright, it’s better if we try to do something else. Yep. AUDIENCE: So the
mentality of walk before you’re ready and
[INAUDIBLE] can sometimes lead to quality loss,
so how do you balance reaching more users
so the question was, launching before you’re
ready and iterating can lead to quality
loss, so how do you balance getting new
customers by launching new features versus pissing
off your existing customers because you break something? It can be a tricky balance. But part of it is in defining
what your problem space is, right? And so defining the things
that you’re going to launch and iterate on. And your customers, if you’re
good to them, and your users, if you’re good to
them and you’ve done good things
for them, they’ll deal with a little
breakage, as long as you don’t break the
core of what they want. If you ever break something
that’s really, really important to them, you’ve got
to be apologetic, and you’ve got to do
good customer service and explain to
them what happened. Google goes down– Gmail goes
down every once in awhile. And it’s crazy, but you
expect that it’s normally up and you know they’re trying
to do good things for you. Same thing happens
with Facebook, although you could argue their
users aren’t– their users are locked in in a different way. But it’s really always a
balance in trying to figure out. In the scope– in the
decision of launch, don’t launch, it’s
basically like if it’s going to get
out there and not– and it’s slightly imperfect,
do it, experiment with it, and be able to roll it back. And roll it back as
soon as you realize it’s hurting your customers and
hurting people who love you. And then do it again and again. Yeah. AUDIENCE: [INAUDIBLE]
customers, how did you know what to focus on?
[INAUDIBLE] 10 customers, you’ve got 20
different opinions. AARON HARRIS: Yeah. So when we first started and
started talking to customers, how did we know
what to focus on. So this is kind of
where– one of the things we tell founders is to work
on problems that you have and things that you know. So I had been a tutor. I had had tutors. So I knew the initial set of
criteria we had to work with and so spared the
first iteration cycle. But then you’ve
got to keep talking and you’ve got to form
consensus to some extent. You don’t want to just go with
exactly what users tell you, because sometimes they’ll
tell different things. It’s a judgment question. There’s no there’s scientific
method to say, do this thing. When Steve Jobs launched
the iPhone, it wasn’t like they had a whole bunch of
user groups who told them, hey, we want a computer in this
form factor that makes calls or whatever. He kind of understood what
people would want eventually. For us, it was constantly
talking to the users, figuring out whether or not
what they wanted matched up with how we thought
the product should grow and ultimately making
product decisions based on our own expertise
on how to fit that advice in. AUDIENCE: Did you
by talking to them or by measuring
the way they acted? AARON HARRIS: Both. Did we measure by
talking to them or by quantitative methods? Both. You have to combine both. There’s pretty much
never an excuse not to talk to your
customers in person. Even when you get
to scale of hundreds of thousands or
millions of users, you can still talk
to pieces of them. This is why community– the
community sites or community forums work really
well as you scale, because you start seeing the
best– the most active users start rising to the top,
and you engage with them, because they are your biggest
fans and your best customers. And if you make
them happy, they’ll spread the love of how
happy they are with you. So you’ve got to do both. Yeah. AUDIENCE: [INAUDIBLE]? AARON HARRIS: So how did we find
our users to do user testing? So initially, the
people that we targeted, it was friends and
friends of friends. But then it was
our actual users. We had people paying us
money to use our service, so we would call them on
the phone– email them, call them on the phone, badger
them, go meet them for coffee, just get– find any way we
could to go talk to them. We never used a
user testing service to find us probable people. We actually focused on
the people that we had. Because we had them, so
it made sense to use them. Yeah. AUDIENCE: [INAUDIBLE]? AARON HARRIS: Yeah,
so I mentioned that we cut a million pages– or
the million pages we had ended up being one of the things
that Google penalized. So we made that optimizat–
and whether or not that was actually
bad for users, that was an optimization
we did– what? AUDIENCE: [INAUDIBLE]? AARON HARRIS: So you’re
saying that’s not product, but I disagree. Part of our product was
the search interface, the browse interface of
how people actually located a tutor. And so the tutors
were distributed across this large
corpus of pages. And how they interacted
with each individual page was the start of their
experience with our product. AUDIENCE: [INAUDIBLE]. AARON HARRIS: So where they
unqualified users because of [INAUDIBLE]? Even if they were
qualified users, they weren’t having the
best possible experience on those pages. Yeah. Yeah. AUDIENCE: [INAUDIBLE]? AARON HARRIS: So how many people
were working and our revenue numbers– so I’ll tell you
how many people were working for us, but– at
our peak, we had, I think, 10 full-time,
5 part-time, and then a whole group of
distributed people doing even more part-time
work– so 15 to 20 people. AUDIENCE: [INAUDIBLE]? AARON HARRIS: I’m taking
it to the grave with me. No, we were generating
good revenue. We were profitable. We were edging on
profitability, growing quickly. It was going well for a while. What? How much money did we raise? We raised one– just shy
of $2 million– just shy of $2 million. Anyone else? Last question. AUDIENCE: Going back to
the [INAUDIBLE] position, did you have a really weird
niche that you were going after and it might not be
your [INAUDIBLE]? AARON HARRIS: So how do you
do user acquisition when your users are not
your circle of friends? It’s like a weird niche? I would go and talk to them. I don’t know. Every niche of users
is going to have slightly different behaviors. Go find them and talk to them. Find out where they hang out,
what they do, what they read. And you can start
building off of that. So, yeah there’s no
science to user– there’s some science
to user acquisition, but the science is in– there’s
no– the science is constantly iterating on and
trying different things and then measuring and then
trying it again and altering it. All right, thank you very much.

, , , , , , , , , , , , ,

Post navigation

One thought on “Google I/O 2014 – How crashing my startup taught me to work at Y Combinator

  1. Actually, your user base dropped because everyone was at my birthday party. Thanks for the interesting perspective and useful insight.

Leave a Reply

Your email address will not be published. Required fields are marked *